The paper introduces BiSeNet V2, a novel architecture designed for real-time semantic segmentation that balances low-level details and high-level semantics. The architecture consists of two branches: the Detail Branch, which captures low-level spatial details with wide channels and shallow layers, and the Semantic Branch, which focuses on high-level semantics with narrow channels and deep layers. A Guided Aggregation Layer is designed to merge the complementary features from both branches, enhancing the overall representation. Additionally, a booster training strategy is proposed to further improve segmentation performance without increasing inference cost. Extensive experiments on Cityscapes, CamVid, and COCO-Stuff datasets demonstrate that BiSeNet V2 achieves a significant trade-off between accuracy and speed, outperforming state-of-the-art methods in both respects.The paper introduces BiSeNet V2, a novel architecture designed for real-time semantic segmentation that balances low-level details and high-level semantics. The architecture consists of two branches: the Detail Branch, which captures low-level spatial details with wide channels and shallow layers, and the Semantic Branch, which focuses on high-level semantics with narrow channels and deep layers. A Guided Aggregation Layer is designed to merge the complementary features from both branches, enhancing the overall representation. Additionally, a booster training strategy is proposed to further improve segmentation performance without increasing inference cost. Extensive experiments on Cityscapes, CamVid, and COCO-Stuff datasets demonstrate that BiSeNet V2 achieves a significant trade-off between accuracy and speed, outperforming state-of-the-art methods in both respects.